The Math Behind a Crazy Idea

Awayday

The quietest line in Awayday's new investment carries the most weight: some of the proceeds are going to the local operators and team members who built the value.

Warburg Pincus LLC has invested in Awayday Vacation Rentals through its Capital Solutions Founders Fund. The amount and terms were not disclosed. Ares Management and LightBay Capital remain the controlling shareholders, which makes this a liquidity event without a disclosed change in control.

Awayday manages more than 18,000 vacation-rental properties across the United States through local brands supported by more than 1,800 people. The shared platform handles technology, revenue management, marketing, distribution, finance, HR and other infrastructure. The local teams keep the names, market knowledge, owner relationships and operating calls that made those businesses worth joining.

Vacation rentals punish the fiction that every market behaves the same. Gulf Coast pricing and a Montana ski week do not share a calendar. A guest locked out at midnight needs someone who knows the home and the market, not a national strategy deck.

Awayday's proposition is that national scale should support that judgment without replacing it.

The company says more than 60 local operators have joined its platform since 2021 and founder-partners still own more than 30% of the business. In May 2025, Awayday reported more than 9,000 properties under management. The October 2026 announcement reports more than 18,000, although the company has not broken out how much of that increase came from acquisitions versus organic growth.

That expansion is impressive. It also increases the operating tension. Each acquired brand brings its own people, processes, homeowner promises and reputation. Central systems have to remove work without removing authority from the people closest to the guest.

Jakob Dwyer, CEO of Awayday, calls ownership mentality a core value. This transaction gives that phrase a financial consequence. Proceeds will return capital to Ares, LightBay and rollover equity partners, while also reaching leading local operators and supporting team members.

The result is a sharper test for Awayday. It has shown that local founders can join a national platform, retain identity and participate in the upside. Now the company has to keep those operators feeling like owners as the platform becomes larger, the capital stack becomes more institutional and the distance between headquarters and a guest's front door keeps growing.

Lambda

AI infrastructure has spent years asking capital markets to finance a promise. Lambda just handed them a schedule.

The company closed a $1.008B delayed-draw, senior secured term loan at a 6.78% fixed rate. Capital is released as GPU clusters enter service, then amortizes after each tranche stabilizes. That detail matters because Lambda is avoiding the cost of borrowing the full amount before the infrastructure can produce revenue.

The loan will support more than 30,000 NVIDIA GB300 and VR200 GPUs across data centers in Seattle, Kansas City, and Dallas. Those systems are tied to 3 committed customer deployments and 2 highly rated counterparties under take-or-pay contracts. Lambda has not named the customers, and it has not disclosed how much of the facility has already been drawn.

Morningstar DBRS assigned the transaction an A (low) rating. Moody's assigned Baa1. The debt is secured by the funded GPU servers, related infrastructure, and contracted cash flows. J.P. Morgan served as sole coordinating lead arranger, structuring agent, and bookrunner.

That structure turns a fast-aging piece of hardware into something fixed-income investors can underwrite like essential digital infrastructure. The silicon still matters. The stronger collateral story is the contract wrapped around it: who pays, for how long, under what service obligations, and how quickly the equipment begins producing cash.

The risks are just as physical. Morningstar DBRS identified GPU delivery and acceptance, unfinished data-center capacity, the short operating history of NVIDIA's GB300 and VR200 systems, and Lambda's own parent-level credit profile during installation. AI may live in the cloud. The debt still has to survive construction schedules, power systems, uptime requirements, and hardware failures.

Michel Combes, CEO of Lambda, is building on a capital stack that already included a $1B bank facility in May and a separate $926M term loan B in August. Founders Stephen Balaban and Michael Balaban now lead technology and product as CTO and CPO. Each facility reaches a different pool of capital, while customer commitments increasingly determine how much infrastructure can be financed.

This is what industrialization looks like before it feels industrial. Compute demand becomes a contract. The contract becomes collateral. The collateral buys GPUs. Then 30,000 machines have to arrive, be accepted, meet their service levels, and turn those financing assumptions into operating cash flow across 3 cities.

Halluminate

The internet is a terrible teacher once the exam looks like an actual job.

A model can explain a leveraged buyout, summarize a contract, and produce a polished slide. Then the instructions change across 21 emails, the data room holds 160 files, and somebody still expects the final deliverable to be right.

That gap is where Halluminate has built its business.

Co-founders Jerry Wu, CEO, and Wyatt Marshall, CTO, started the San Francisco company in 2024 to build reinforcement-learning environments for AI agents doing knowledge work. Halluminate has now raised a $30M Series A led by Oak HC/FT, with Y Combinator, Orange Collective, FT Partners, and Heavybit participating. The company says total capital raised is now $38.5M.

Halluminate recreates the workplace around the task: spreadsheets, documents, email, chat, data rooms, changing instructions, and verifiers that can tell whether the work survived the process.

Its Westworld Finance Diligence Bench uses 88 tasks drawn from anonymized private transactions and reviewed by practicing deal professionals. Agents move through modeling, underwriting, pitch materials, closing, and diligence work that can stretch across hundreds of steps.

The strongest tested configuration earned roughly half the available credit. Fluency gets attention. Reliability gets the mandate, especially when one missed instruction can corrupt the model, the memo, and the decision built on top of both.

Halluminate reports a mid-eight-figure annualized revenue run rate in 10 months, profitability, and work with four of the five leading closed-source U.S. AI labs. Those are company-reported metrics, but they explain why Oak HC/FT is underwriting a specialized data research lab while the market still argues about how much intelligence fits inside a benchmark score.

The founders' wager is that the next valuable training data will be engineered from real workflows by people who understand which details matter, then made difficult enough to keep teaching models that improve every few months.

That creates Halluminate's own unforgiving assignment. It has to expand beyond finance without losing domain depth, keep its environments ahead of the agents learning inside them, and turn a concentrated customer base into durable infrastructure.

The $30M buys more room to work on that assignment. The interesting part is how quickly Halluminate can make professional judgment teachable before the models arrive for the next lesson.

SignSplit

AI keeps getting better at turning human work into training data. It remains much worse at proving who agreed, who owns what, and who gets paid.

SignSplit is building for that gap. The company emerged from stealth with a $400M strategic seed financing commitment from W Group at a stated $1B valuation.

That wording matters. The package combines capital with multi-year strategic resources, and the parties did not disclose the cash split, funding schedule, or value assigned to those resources. This is a very large commitment. It is not evidence that $400M in cash arrived on day one.

Co-founder and CEO Alessandro Monterosso and co-founder, Executive Chairman and CPO Glib Denisov call the model “signed data.” People and institutions can protect, license, and contribute data, work, and likeness under defined consent and usage terms. AI companies, robotics developers, researchers, and media platforms can request data or research pools built around specific needs, while contributors can be compensated for participating.

Most data-rights fights begin after the data has already moved. A creator finds a work inside a model. A company discovers that a purchased dataset cannot explain who consented. A research team needs longitudinal human behavior that nobody collected because the request did not exist yet.

SignSplit wants the permission, provenance, and payment record to exist before the handoff.

W Group gives that bet a strategic shape. The ecosystem led by Volodymyr Nosov reports 40M users worldwide and operations across 15 locations. Distribution, payment infrastructure, and cross-border operating reach may matter as much as the headline capital when the product depends on networks of buyers and contributors.

The opening is substantial. So is the proof still required. SignSplit has not publicly disclosed customers, revenue, deployments, or how the verification layer performs when data moves through collection, model development, and commercial use.

The next useful numbers will be active data pools, repeat buyers, contributor retention, verified payments, and rights that survive procurement. That record also has to survive deletion requests, licensing changes, payment disputes, and a buyer's own compliance review. SignSplit has put an extraordinary commitment behind the idea that human contribution should carry its own commercial memory. Now that memory has to stay attached while every system around it is designed to copy.

PTC

Industrial AI has a memory problem. The most important decisions about a machine are usually made before it starts producing operating data.

Engineering teams decide the geometry, requirements, software behavior, service plan, and failure modes. Operations inherit the telemetry, energy demands, and consequences when that design intent becomes hard to recover.

Schneider Electric is proposing to own more of that handoff.

The company agreed to acquire PTC for $205 per share in cash, valuing PTC's equity at about $22.6B and implying a $23.7B enterprise value. That is a 42.3% premium to PTC's October 2 closing price. Closing is expected by Q3 2027, subject to shareholder and regulatory approvals.

PTC holds product memory across Creo, Onshape, Windchill, Arena, Codebeamer, ServiceMax, and Servigistics. Together they connect design, product data, software requirements, and service.

Schneider sits closer to the physical world: energy, automation, factories, infrastructure, and, through AVEVA, operational software. Its proposed Cognite acquisition adds industrial data and AI context.

PTC fills the upstream gap. It brings the record of what a product was supposed to become before a sensor reports what it actually did.

That is the strategic wager behind the price. Olivier Blum is trying to connect engineering intent with operating reality and give industrial AI trusted context across design, build, operation, and maintenance. If the transaction closes, PTC's Intelligent Product Lifecycle would gain access to Schneider's larger global channel and deeper position in energy and automation.

The financing makes the conviction visible. Schneider expects roughly EUR5B-EUR6B of new equity and EUR16B-EUR17B of new debt. Management projects EUR250M in annual run-rate cost synergies by year 3 and approximately EUR800M in revenue synergies. Those remain forecasts, and the acquisition still has to clear shareholders, regulators, and the less ceremonial work of integration.

Customers will care about the word Schneider keeps repeating: open. Industrial environments run on mixed hardware, old software, acquired systems, and processes that refuse to fit a clean architecture diagram. A unified digital thread has value only if it crosses those boundaries without turning every customer into a captive integration project.

If the transaction closes, Schneider would acquire the decisions made before the dashboard had anything to measure, betting that industrial AI becomes more valuable when the product and the plant can remember the same story.

Serán Bioscience

Commercial scale begins with an uncomfortable request: hand a drug formula to equipment, operators and a quality system that did not watch it grow up.

The equipment changes. The batch grows. A new team inherits decisions it did not make. Quality systems, packaging, serialization and distribution all enter the picture while the medicine is supposed to keep behaving exactly as intended.

Serán BioScience is raising capital around that moment. The Bend, Oregon CDMO announced an undisclosed growth investment led by Biospring Partners, with participation from Vivo Capital, Bain Capital and Serán's executive leadership.

Serán is building a multi-suite commercial manufacturing facility scheduled for completion in Q3 2027. The site is designed for OEB4 manufacturing across oral, pulmonary and other formats, with spray drying, nano-milling, granulation, tableting, coating, encapsulation and dry-powder filling.

It will connect with an adjacent packaging and logistics facility that Serán says can handle validated packaging, labeling, unit-level serialization and global cold-chain distribution. Together, the expansion is expected to create a Bend campus exceeding 200,000 square feet.

That footprint matters because a CDMO customer is rarely buying equipment alone. The customer is buying fewer transfers of knowledge between the scientists who designed a formulation, the engineers who scale it, the operators who manufacture it and the teams responsible for getting finished product where it needs to go.

Serán was founded in 2016 by Dan Smithey, Ph.D., and John Melvin. Dan Smithey remains president and CEO. The company now reports more than 200 employees, including over 150 scientists and engineers, and expects to add as many as 150 people as the commercial expansion moves forward.

Bain Capital Life Sciences led a separate strategic growth transaction of more than $200M in 2024, with Vivo Capital remaining an investor. Biospring is joining that group around demand for complex particle engineering and commercial manufacturing.

The unfinished work sits between construction and dependable supply. Serán still has to complete the site, validate the operation, hire and train the next wave of employees, qualify customer programs and reproduce development-stage knowledge at commercial scale.

For drug developers, that is the value proposition taking shape in Bend: fewer institutional seams at the point where a promising formulation becomes a manufacturing obligation.

UpSmith

A home-services company can be drowning in demand and still leave revenue sitting untouched inside its own database.

The missed call came in after hours. The estimate never got a second follow-up. A maintenance agreement expired. A loyal customer quietly disappeared from the schedule. Across thousands of records, small lapses become an expensive habit.

UpSmith has raised a $10M Series A to work that neglected layer of the business. SemperVirens VC led the round, joined by Hannah Grey VC, Asymmetric Capital Partners, CIV, Synetro Group, WaterStone Impact Fund, and other investors.

Founder and CEO Wyatt Smith started UpSmith in 2022 around the skilled-labor shortage. The company has since evolved from workforce development into an AI platform for HVAC, plumbing, and electrical contractors. That shift is less of a departure than it first appears. A contractor cannot create more skilled workers overnight, but it can stop wasting the demand, time, and customer history its existing team already earned.

UpSmith's agents answer inbound calls, respond to time-sensitive leads, confirm appointments, follow up on open estimates, renew memberships, and re-engage inactive customers. The company is also introducing a voice agent and a unified customer timeline designed to turn those interactions into one operating record.

The results are company-reported. UpSmith says it serves more than 200 home-services operators. It reports supporting more than 5,000 completed jobs from recovered estimates for HomeTown Services in 2026, reducing cancellations by 34% at Five Star Home Services, and helping Medley AC generate $147K in new revenue over 3 months.

Those numbers explain why the round is not merely another wager on an AI assistant. Contractors already have software, phones, customer records, and people who know the work. Their problem is that too many revenue-producing moments still depend on somebody having enough time to notice and act.

The capital will expand UpSmith's engineering team and distribution. Allison Baum Gates of SemperVirens and Rob Biederman of Asymmetric are joining the board, putting investor operating help closer to the company's next stage.

That is the pressure inside this round. UpSmith is selling capacity to businesses whose reputation still arrives in a truck, at a customer's door, when something important has stopped working. If its software can keep more of that demand from going cold, the AI disappears into the schedule and the technician gets another job worth showing up for.

Clockwork.io

One failed link can turn a room full of healthy GPUs into very expensive spectators.

Clockwork Systems, Inc. raised $31M to keep that audience from forming. Premji Invest, Wing Venture Capital and Seligman Ventures co-led the round, with NEA and e& Capital returning. Total funding is now $73M.

The amount matters because of the operating problem underneath it. Distributed AI training can stretch across thousands of accelerators that have to move together. A failed GPU, NIC, cable, switch port or server can stall the job, leave healthy hardware waiting and force a restart from an earlier checkpoint. Meta documented 419 unexpected interruptions during a 54-day snapshot of Llama 3 training on 16,384 H100 GPUs, roughly 1 every 3 hours.

That creates an infrastructure problem with a meter running. Suresh Vasudevan leads Clockwork.io with a systems team rooted in Stanford research. Co-founders Balaji Prabhakar and Deepak Merugu built the company with CTO Yilong Geng around software timing precise enough to observe distributed systems. Clockwork turned that foundation toward keeping expensive AI work moving when a large cluster inevitably fails somewhere.

LinkPass reroutes traffic around a failed network path. TorchPass moves work from a failing GPU to a healthy one rather than rolling the entire job back. The new release adds multi-node snapshots that preserve a distributed job across every node and asynchronous checkpoints designed to move updated model weights faster during reinforcement learning.

The customer evidence is becoming the story. LinkedIn says Clockwork.io prevents tens of thousands of GPU-hours of downtime each month. Together AI is bringing TorchPass to market on its GPU clusters. WhiteFiber is expanding the software across its GPU-as-a-service footprint. Those are customer and company-reported results, but they put the product inside real operating consequences instead of a resilience demo.

The three new co-leads are backing a shift in how AI infrastructure gets judged. Buying capacity is visible. Recovering useful work when something breaks protects the return.

Clockwork.io will use the capital to expand fault tolerance across training, inference and reinforcement learning, deepen enterprise adoption and scale through cloud partners. That expansion now has to preserve the same thing its software promises customers: progress that survives the next failure without making every healthy GPU wait for the recovery plan.

Valon

For six years, Valon did the job most software companies prefer to study from a safe distance. It operated a mortgage servicer.

That meant payments, escrow, defaults, investor reporting and the edge cases that appear after the demo ends. Every exception became a product requirement. Every audit exposed whether the system understood what happened and why.

Now Valon has raised a $150M Series D at a $2.3B valuation. Ribbit Capital joined as a new investor, while Andreessen Horowitz returned. The money will accelerate product development and expand teams as Valon moves more major servicers off legacy systems.

The round matters because AI inside regulated finance has a context problem. A model can draft an answer quickly. Mortgage servicing still needs to know which payment moved, which rule applied, what changed in escrow and how the decision will look to an examiner months later. Intelligence without that record is a fast way to create slower investigations.

ValonOS puts loan data, accounting ledgers, investor reporting, workflows, compliance logic and money movement in one system of record. Ditto, its native agent layer, can then handle work such as homeowner communications, escrow analysis, complaint research and payment reallocations under the same controls.

The company says the platform has crossed $200B in mortgages, signed more than $200M in contracted ARR within six months of opening ValonOS to outside servicers, and has customers responsible for 1 in 6 outstanding U.S. mortgages under contract. Carrington and ServiceMac are live. Newrez plans to begin its transition in 2027.

Those are company-reported numbers, and the hard part now moves from product proof to enterprise implementation. Replacing core servicing technology means migrating decades of data, process and institutional memory without giving homeowners or regulators a reason to notice the handoff.

CEO and co-founder Andrew Wang, President, COO and co-founder Linda Du, and co-founder Jonathan Hsu took the long route because the shortcut could not teach the software enough. Carrington's acquisition of Valon Mortgage completed that first chapter: the operating business proved the platform, then the technology company stepped forward.

The $150M gives Valon more room to turn that proof into infrastructure for the industry's largest servicers. The next work happens inside the migrations, where every exception the old system learned over decades has to arrive intact before an agent can act with confidence.

Consano Bio

Chronic sciatica can turn a damaged nerve into a daily negotiation with pain, mobility, work and sleep. Consano Bio raised more than $15M to test whether medicine can change the damaged biology instead of negotiating around it.

The oversubscribed private financing came from unnamed accredited investors. Consano Bio says it has raised more than $45M since 2023. The capital has a specific job: complete a 24-patient Phase 1 study of C-1101 and keep building the science and IP estate around the program.

C-1101 is a platelet-derived, multi-protein biologic delivered by epidural injection. It is designed to place cytokines, growth factors and matrix proteins at the injury site to modulate inflammation and promote localized repair.

Chronic sciatica is often managed with treatments aimed at temporary or incomplete relief. Consano Bio is building a pharmaceutical therapy around the underlying biology of injury. The ambition is disease modification. The evidence is still being built.

Andrew Hall, CEO and board member, and Jennifer Schmitke, PhD, COO and Head of R&D, have moved the program into a randomized, controlled, double-blind dose-escalation study in Australia and the United States. Each participant receives one injection. Enrollment is expected to finish in early 2027, with clinical data expected in 1H 2027.

The regulatory path has also tightened. FDA cleared Consano Bio's IND for U.S. expansion in July and granted C-1101 Fast Track designation in August. Fast Track can increase communication and speed parts of development and review. Approval still depends on the clinical evidence, including the question patients care about.

The preclinical record gives the team a reason to ask that question. A peer-reviewed September paper reported reduced pain sensitivity in a mouse model and biological activity in cell models. It also found a mixed inflammatory response and called for more work on the mechanism. The study was funded by Consano Bio, with company employees among the authors.

That is the useful shape of this financing. Private investors are funding the handoff from a scientific argument to controlled human evidence. The headline says more than $15M. The next meaningful numbers belong to safety, tolerability and the patient data that follow.

Consano Bio expects that readout in the first half of 2027. This financing gives the company runway to reach it, then lets 24 patients begin telling the part of the story the lab cannot finish.

Namespace

A coding agent can create another commit before the previous one has finished testing. That sounds like leverage until the build system becomes the most expensive person in the room, billing engineers and agents in time.

Namespace has closed a $42M Series B led by Scale Venture Partners, with New Enterprise Associates (NEA), 20VC, Essence, Burst Capital, Susa Ventures, and Datadog CEO Olivier Pomel participating. The round arrives seven months after Namespace disclosed $23M across Seed and Series A financing, bringing total funding to $65M.

Founder and CEO Hugo Santos is building around a simple mismatch. Software generation is becoming cheap and abundant. Building, testing, reproducing, and securing that software still requires real machines and the right operating system at burst scale.

Namespace provides coding, build, and test environments across Linux, Windows, and Mac, with Devboxes for coding agents and a drop-in replacement for GitHub-hosted runners. Underneath sits infrastructure Namespace designs and operates itself, from compute and storage to server racks deployed around the world.

The Mac piece is easy to dismiss until an agent needs to compile and test an iOS application. A cloud agent can write the code anywhere, but proving it works on macOS or inside an iOS simulator requires Apple hardware and tooling many infrastructure providers skip.

Namespace says revenue grew 8x over the past 12 months and that more than 1,000 companies use the platform, including Ramp, Framer, Vanta, and Warp. Those are company-reported numbers, but the investor logic is visible: code volume may grow much faster than the engineering capacity available to nurse brittle pipelines.

The new capital will accelerate product development and expand Namespace's data centers, including its Mac fleet. That puts a physical obligation underneath a software thesis. Every new rack turns performance from a benchmark into an operating promise that customers will test on their busiest day. Custom hardware and owned infrastructure can create performance and control, but they also demand deployment discipline, utilization, reliability, and customer growth that keeps expensive capacity useful.

The next 100 billion commits will not be valuable because agents can produce them. They become valuable when somebody can build them, test them, reproduce them, and decide which ones deserve to reach production. Namespace is financing the machinery behind that decision.

RXO

Freight brokerage looks asset-light until a load goes sideways.

Then the asset becomes visible: the carrier that answers, the price history that catches a bad quote, and a network dense enough to offer another way through.

C.H. Robinson has agreed to acquire RXO, Inc. in a cash-and-stock transaction with an implied value of $5.8B. The deal would combine two major North American brokerage networks inside a company with an enterprise value above $25B.

Standard consideration is $17.25 in cash plus 0.0856 C.H. Robinson shares per RXO share, implying $30.25 using the deal's reference price. Proration is designed to leave the aggregate mix about 57% cash and 43% stock, with RXO holders owning roughly 11% of the combined company.

Announced October 5, the transaction is expected to close in the first half of 2027, subject to RXO stockholder approval, regulatory clearance, and other conditions. This is an agreement, not a finished acquisition.

The logic lives deeper than market share. C.H. Robinson brings a global multimodal platform that managed about 37M shipments for 75,000 customers in 2025. RXO adds truck brokerage, managed transportation, expedited freight, and a last-mile operation that arranges more than 11M deliveries a year.

That puts more freight, carriers, lane history, last-mile capacity, and data inside one system. For a shipper, useful scale means finding capacity, recovering from disruption, and moving between modes without turning the handoff into a scavenger hunt.

C.H. Robinson expects to integrate RXO primarily into its North American Surface Transportation division and apply its Lean AI operating model. Management projects approximately $300M in annual net run-rate cost synergies within two years after closing and expects the larger data estate to improve sales, matching, and procurement.

Those are forecasts. Integration will decide whether they become facts. Dave Bozeman is committing C.H. Robinson to new debt and a pause in share repurchases until leverage returns to its target range. If the transaction closes, Drew Wilkerson would move RXO, born from XPO's 2022 spin-off and expanded through the $1.025B Coyote acquisition, into a larger system.

The work now sits in ordinary freight decisions: which systems survive, who owns the exceptions, whether customers feel more capability or more distance, and whether network density makes the next load easier to cover. A $5.8B signature creates the map. The combined operators still have to move the freight across it.

Ocular AI

The most revealing moment in a voice conversation may be the half-second nobody writes down. Someone interrupts, the other person pauses, a quiet “mm-hmm” keeps the thought moving, and the meaning changes without changing the transcript.

Ocular AI is building for that missing layer. The San Francisco company announced a $2M pre-seed round led by Drive Capital, with Y Combinator, Alumni Ventures, 1745 Ventures, Orange Collective, MyAsia VC, and angel investors participating.

CEO and co-founder Michael Moyo and CTO and co-founder Louis Murerwa started Ocular inside Y Combinator’s Winter 2024 batch. The company first worked on enterprise search and cross-tool actions. Its current wager reaches deeper into the model stack: build an applied AI data research lab that can turn human expertise into training data, evaluations, and benchmarks for frontier voice and multimodal systems.

That distinction matters because clean data can make a model look more capable than the customer experience feels. Real conversations contain accents, overlap, false starts, corrections, emotion, facial expression, and the split-second judgment required to know whether to speak, wait, or stop. A transcript preserves the words and can still lose the conversation.

Ocular combines a Data Foundry with a network it says includes thousands of vetted domain experts. The company captures full-duplex speech on separate high-fidelity tracks, builds audiovisual datasets, and creates evaluations that show where models break. Its Converse-STT benchmark compared 15 speech-to-text models on real two-person conversations and public Pipecat audio. Ocular reported that 12 performed worse on the conversational data.

Ocular says revenue has reached seven figures and that its datasets and evaluations are used by unnamed frontier AI labs and Fortune 100 enterprises. Those are company-reported claims, not audited customer disclosures. They still explain why investors are backing the move from enterprise search toward the data and evaluation layer.

The round gives Ocular room to expand Converse, hire engineers and researchers, and move from voice into audiovisual systems and professional domains. Growth will test whether expert quality, consent, licensing, and provenance survive volume.

The internet gave AI more language than any person could read. Ocular is betting that the next advantage comes from the human detail the internet never captured cleanly enough to teach.

OneByZero

Niket Vaidya and Vibhore Kumar spent three years choosing the least glamorous part of enterprise AI: staying after the demo.

A model can impress a room in an afternoon. Production has to survive the bank's permissions, the telecom's call queues, the retailer's legacy systems, and the meeting where somebody asks who is accountable when an agent takes the wrong action.

The Singapore company has raised a $20M Series A led by Jungle Ventures, its first external financing. The capital will expand forward-deployed teams across nine markets, establish a local team in Japan, enter conglomerates, healthcare, and the public sector, and accelerate NEO, OneByZero's platform for deploying and governing AI Coworkers.

Engineers work beside enterprise teams to identify a valuable workflow, connect AI to the systems already running the business, move it into production, then stay involved as the system and business change.

NEO is meant to turn that close-in work into infrastructure. OneByZero says the platform gives customers role-based permissions, audit trails, escalation rules, kill switches, evaluation loops, and human accountability inside their own AWS environment.

That combination explains the investor logic. Yash Sankrityayan said OneByZero stood out because the team was already doing what much of the market still discusses in theory: deploying AI inside large, complex organizations.

OneByZero says revenue has more than doubled annually for three years, some customer-facing deployments automate more than 90% of interactions, and complex data-modernization work has moved 50% faster. Those are company-reported figures, not audited financial results.

Forward-deployed engineering can create trust, proprietary operating knowledge, and systems that fit the customer's actual constraints. It can also become a sophisticated services business unless each deployment makes the next one easier to build, govern, and sell.

That is why Japan and the investment in industry templates matter. Niket Vaidya and Vibhore Kumar are using the round to add people near customers while asking NEO to preserve more of what those people learn.

The companies moving enterprise AI from pilots into daily work will earn their place one permission, escalation path, integration, and accountable handoff at a time. OneByZero now has $20M to see how much of that work can compound across markets without losing the local judgment that got it into production.